Professional Documents
Culture Documents
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Oxford University Press is collaborating with JSTOR to digitize, preserve and extend access to The Quarterly
Journal of Economics.
http://www.jstor.org
I. INTRODUCTION
1128
QUARTERLYJOURNAL OF ECONOMICS
1129
1130
QUARTERLYJOURNAL OF ECONOMICS
the model, where the saving rate is dynamically optimized, also has
these implications.
It has now been quite some time that researchers have been
confronting real data with these hypotheses. Initially, much of this
work was conducted on the basis of the data of the developed
industrialized countries. Data availability had a significant role in
this choice of sample. In one of the recent works on this topic,
Baumol [1986], for example, reported finding convergence among a
group of countries included in Maddison's [1982] sample. These
countries tended to converge both to similar levels of per capita
income and to similar rates of growth.
An important question in this regard is what should be the
appropriate methodology for testing convergence. Since the notion
of convergence pertains to the steady states of the economies, a test
for convergence would require the assumption that the countries
included in the sample are in their steady states. However, judging
whether countries are in their steady states or not can be problematic. One way around this problem, therefore, is to study the
correlation between initial levels of income and subsequent growth
rates. Because of diminishing marginal returns to capital, countries with low levels of capital stock will have higher marginal
product of capital and hence, for similar saving rates, grow faster
than those with already higher levels of per capita capital stock.
Thus, a finding of negative correlation between initial levels of
income and subsequent growth rates has become a popular criterion for judging whether or not convergence holds. It may be noted
that this negative correlation has the scope of being interpreted as
evidence of convergence in terms of both income level and growth
rate. Poorer countries "catch up" (convergence in terms of income
level) with the richer countries by initially growing faster, and then
their growth rates slow down to the common rate of technological
progress (resulting in convergence in growth rates).
As more wide-ranging data sets became available, empirical
regularities of the growth process over a wider cross section of
countries started to draw the attention of researchers. Romer
[1989a] has been influential in drawing the attention of macroeconomists to the fact that over a large sample of countries, the
correlation between initial income levels and subsequent growth
rates is either zero or even positive. The evidence has also been
interpreted as one of "persistence" of significant differences in
income level and growth rates among countries [Rebelo 1991; King
1131
1132
QUARTERLYJOURNAL OF ECONOMICS
Y(t) = K(t)0(A(t)L(t))1-a
0 < a < 1,
1133
Y(t)1
in =
lnA(O) + gt +
In (s) +-
In (n
g +
E,
1134
My
In \L=a
aa
+ 1_
In (s
1-(
In (n
+g
+ 8)
+ e.
1135
term E.This is better revealed by considering the equation describing out of steady state behavior. The way this equation is derived is
as follows. Let 9* be the steady state level of income per effective
worker, and let 9(t) be its actual value at any time t. Approximating
around the steady state, the pace of convergence is given by
dIn9(t)
(5)
9(t)],
X[ln (9*)-In
dt-
dt
where A = (n
(6)
In9(t2)
(1
e-AT)
e-AT)
In9(t1).
This equation represents a partial adjustment process that becomes more apparent from the following rearrangement:
In9(t2) - In 9(t1) = (1 - e- XT)(In9* - In9(t1)).
(8)
9* gives
(9)
-(1
e XT)
ln
1-(x
In (s)
(n +g+5)
- (1 -e-XT)ln9(tl).
QUARTERLYJOURNAL OF ECONOMICS
1136
Y(t)
L(t)A(t)egt
Y(t)
9(t)
A(t)L(t)
so that
IY(t)\
I - In A(O) - gt
In9(t) = In I\
L(t)/
= In y(t) - In A(O) - gt,
where y(t) is the per capita income, [Y(t)/L(t)]. Substituting for 9(t)
into equation (9), we get the usual "growth-initial level" equation:
(10)
n (s)
I_
l (n + g + ) -(1
In
- e-AT)
+ (1 - e
- e-AT)lny(t1)
AT) In A(O) +
g(t2 -e
-Tt)
lny(t2) = (1 - e-AT) 1
- (1
e-AT)
ln (s)
+g + 8) + e -Tlny(t1)
aIn(n
+ (1 - e
AT)
InA(O) + g(t2 -e
-Ttl
(12)
where
yit=
yist-l =
y=
lny(t2)
lny(t1)
e-XT
e~
-(
1137
-XeT)
I2 = -(1 - e-AT)
1-
il = In (s)
x2= In (n + g +)
ki = (1 It
= g(t2
e-AT)
-
In A(0)
XTt )I
and vit is the transitory error term that varies across countries and
time periods and has mean equal to zero. Panel data estimation of
this equation now provides the kind of environment necessary to
control for the individual country effect.
It is clear that this panel data formulation is obtained by
moving from a single cross-section spanning the entire period
(1960-1985) to cross sections for the several shorter periods that
constitute it. We may note that equation (11) was based on
approximation around the steady state and was supposed to
capture the dynamics toward the steady state. It is, therefore, valid
for shorter periods as well. Also, it may be noted that in the single
cross-section regression, s and n are assumed to be constant for the
entire period. Such an approximation is more realistic over shorter
periods of time. The panel data setup allows us, after controlling
for the individual country effects, to integrate this process of
convergence occurring over several consecutive time intervals. If
we think that the character of the process of getting near to the
steady state remains essentially unchanged over the period as a
whole, then considering that process in consecutive shorter time
spans should reflect the same dynamics. However, controlling for
the unobservable individual country effects will create a cleaner
canvas for the relationship among the measurable and included
economic variables to emerge.
IV. ESTIMATION
ISSUESANDDATA
A. Relevant Issues of Panel Estimation
A host of methods is available for the estimation of panel data
models with individual effects.1 One common issue that arises in
1. For a recent review see Islam [1991].
1138
QUARTERLYJOURNAL OF ECONOMICS
->
1139
1140
QUARTERLYJOURNAL OF ECONOMICS
1141
DEPENDENT VARIABLEIs
NONOIL(96)
Sample:
INTER(74)
In (y85)
OECD(22)
Unrestricted
Constant
In (Y60)
In (s)
In (n + g + 5)
0.9448
1.1075
1.7433
(0.8724)
0.8733
(0.0611)
0.6585
(0.0926)
(0.8975)
0.7822
(0.0667)
0.6431
(0.1121)
(1.2655)
0.6722
(0.0694)
0.4114
(0.1845)
-0.6122
(0.3667)
0.9006
0.00542
(0.00037)
R
ImpliedX
-0.8144
(0.3717)
0.8915
0.009827
(0.00083)
-0.8021
(0.4187)
0.8499
0.015887
(0.00164)
Restricted
Constant
In (Y60)
In (s) - In (n + g + 5)
R
Implied X
Implied a
0.8475
1.4565
2.6689
(0.3429)
0.8701
(0.0547)
(0.3798)
0.7945
(0.0599)
(0.5715)
0.6817
(0.0678)
0.6554
0.6610
0.4847
(0.0884)
0.9037
0.005565
(0.00035)
0.8346
(0.1126)
(0.1034)
0.8927
0.009204
(0.00069)
0.7628
(0.1193)
(0.1602)
0.8524
0.015327
(0.00152)
0.6036
(0.1995)
0.95
0.90
0.70
QUARTERLYJOURNAL OF ECONOMICS
1142
TABLE II
POOLED REGRESSIONFROMA PANEL OF FIvE-YEAR SPAN DATA:
DEPENDENT VARIABLEIs Yit
NONOIL
480
Sample:
No. of obs.
INTER
370
OECD
110
0.9636
(0.0107)
0.1396
(0.0172)
-0.1300
(0.0566)
0.9861
0.0074
(0.0001)
0.9228
(0.0147)
0.1047
(0.0313)
-0.1799
(0.0653)
0.9807
0.0161
(0.0003)
0.9628
(0.0098)
0.1388
(0.0165)
0.9861
0.0095
(0.0002)
0.7736
(0.0924)
0.9248
(0.0147)
0.1184
(0.0286)
0.9901
0.0146
(0.0002)
0.6150
(0.1486)
Unrestricted
0.9764
(0.0101)
0.1386
(0.0153)
-0.1291
(0.0584)
0.9848
0.0048
(0.0001)
in (yi,t-i)
in (s)
in (n + g +5)
R2
Implied X
Restricted
in (yi,t-1)
in (s)-in
(n + g +5)
R2
Implied X
Implied a
0.9758
(0.0012)
0.1381
(0.0151)
0.9848
0.0059
(0.0001)
0.8338
(0.0912)
0.90
0.90
0.90
1143
Table I. The implied values of a obtained from these two regressions are also found to be strikingly similar.
These results, therefore, show that dividing the period into
shorter spans and considering the growth process over shorter
consecutive intervals does not affect the results. Both the single
cross section and the pooled regression produce very similar
results. We find very low estimates of the rate of convergence
(particularlyfor the NONOIL and INTER samples) and very high
estimates of the elasticity parameter a. We next see how panel
estimation changes these results.
C. Minimum Distance: Estimation with "Correlated Effects"
,ui = Ko + K1Xij
with E [i lxi,, . .
., XiT]
+ KTXiT + PJi
+ K2Xi2 +
Yio = No + (lXil
+ (2Xj2 +
...
+ (TXiT + Pi,
1144
QUARTERLYJOURNAL OF ECONOMICS
(15)
YU =
Yist-l +
exit
mt+ ki
+ Vit,
where
ax
=
(1
e-XT)
1145
TABLE III
MINIMUM DISTANCE ESTIMATIONWITH CORRELATEDEFFECTS:
DEPENDENT VARIABLEIS yit
Sample
y
0
Implied X
Implied at
NONOIL
INTER
OECD
0.8050
(0.0306)
0.1530
(0.0274)
0.0434
(0.0076)
0.4397
(0.0614)
0.8117
(0.0284)
0.1389
(0.0243)
0.0417
(0.0070)
0.4245
(0.0524)
0.7155
(0.0098)
0.1203
(0.0264)
0.0670
(0.0026)
0.2972
(0.0433)
fact, the relative increase in the impliedvalue of Ais the highest for
the NONOIL sample (by a factor of about 8). Another change that
results from panel estimation is that the impliedvalues of Xfor the
NONOIL and INTER samples are closer to each other. While on
the basis of the single cross-section results, the implied value of A
for INTER was almost twice that for NONOIL, corresponding
values from MD estimation do not differby that much.
Significant change is also observedin the implied values of the
elasticity parameter, at.The estimated value of a for the NONOIL
sample is 0.4397, for INTER, 0.4245, and for OECD,0.2972. These
may be contrasted with the corresponding single cross-section
estimates: 0.8346, 0.7628, 0.6036, respectively. The across-sample
pattern of variation of the parameter estimates remains the same
as obtainedfrom single cross-section regression. The impliedvalue
of the elasticity parameter is found to be the highest for the
NONOIL sample and lowest for the OECD sample. But the panel
estimates for all the samples are much lower and much closer to
their generally accepted values. It is also interesting to note that
the estimated values of a that we obtain from panel estimation are
very close to the estimates that M-R-Wobtainedafter includingthe
human capital variable.
D. LSDV:Estimation with "FixedEffects"
Very similar results are obtained even if one assumes that the
individualcountry effects are fixed in nature. This can be seen from
the results of the LSDV estimation presented in Table IV. Again,
focusing on the estimates of the structural parameters,we see that
1146
QUARTERLYJOURNAL OF ECONOMICS
TABLE IV
LSDV ESTIMATIONWITH FIXED EFFECTS: DEPENDENT VARIABLEIs Yit
Sample:
No. of obs.
Unrestricted
Yit-1
In (s)
In (n + g + 8)
R2
ImpliedX
Restricted
Yit-1
In (s) - In (n + g + 8)
R2
ImpliedX
Implied
NONOIL
480
INTER
370
OECD
110
0.7762
(0.0353)
0.1595
(0.0237)
-0.4092
(0.1024)
0.7404
0.0507
(0.0091)
0.7935
(0.0388)
0.1709
(0.0256)
-0.2466
(0.1007)
0.8254
0.0462
(0.0098)
0.5864
(0.0532)
0.1215
(0.0586)
-0.0698
(0.1007)
0.9659
0.1067
(0.0181)
0.7919
(0.0349)
0.1634
(0.0238)
0.7368
0.0467
(0.0088)
0.4398
(0.0545)
0.7954
(0.0387)
0.1726
(0.0254)
0.8251
0.0458
(0.0097)
0.4575
(0.0575)
0.6294
(0.0495)
0.0954
(0.0581)
0.9642
0.0926
(0.0157)
0.2047
(0.1042)
0.70
0.90
0.90
Figures in the parentheses are standard errors, and the goodness of fit measures are with respect to the
within regression.
1147
A. Statistical Interpretation
A = (1/X) In (,y).
a = /(1-+
13).
1148
QUARTERLYJOURNAL OF ECONOMICS
have seen above what the likely effect on the estimate of y can be
when the correlatedindividualcountry effect is ignored.
We may apply similar reasoning to assess the likely effect of
ignoring the individual effect on the estimate of P. Recall that in
the restricted form, P3is the coefficient of [In (s) - In (n + g + 6)].
The theoretical sign of the coefficient is positive. However, it is
difficult to be sure about the partial correlation of A(O) with
[In (s) - In (n + g + 6)]. Whatever may be the direction of bias in
the estimate of P3,because of its presence in both the numerator
and the denominator,the effect of this bias may, to a large extent,
cancel out, and bias (and its correction)in the estimate of y may be
a determining factor. As equation (18) shows, a lower value of 'y
leads to a lower value of a.
This is, obviously, the narrow statistical interpretation of our
results. We next turn to the interpretation and implications of the
results from growth theory's point of view.
B. Estimation Results and Growth Theory
1149
1150
QUARTERLYJOURNAL OF ECONOMICS
e-AT)
eXT)
ax
In (h*) + e -T In y(tl)
+ (1 - e
AT)
In A(0) + g(t2 -
eXTti)
where h* is the steady state level of human capital, and (pis the
exponent of the human capital variable in the augmented production function of M-R-W.
Inclusion of the human capital variable in our analysis,
however, requires some modification of the sample sizes. This is
because the panel data on HUMANin Barroand Lee [1993] are not
availablefor all the countries of our original samples. Accordingly,
the NONOIL sample now reduces to 79 countries,4and INTER to
67.5 Size of the OECDsample, expectedly,remains unchanged.
4. The countries dropped from the original NONOIL sample are Angola,
Benin, Burundi, Cameroon, Central African Republic, Chad, Congo, Egypt, Ethiopia, Ivory Coast, Madagascar, Mali, Mauritania, Morocco, Nigeria, Rwanda, and
Somalia.
5. The countries dropped from the original INTER sample are Cameroon,
Ethiopia, Ivory Coast, Madagascar, Mali, Morocco, and Nigeria.
1151
TABLE V
ESTIMATIONWITH HUMAN CAPITAL
Variable
Single
cross section
Pooled
regression
Panel
estimation
0.0093
(0.0146)
0.0069
(0.0025)
0.8013
(0.0534)
0.0544
(0.1020)
-0.0712
(0.0323)
0.0375
(0.0093)
0.5224
(0.0642)
-0.1990
(0.1097)
-0.0014
(0.0209)
0.0079
(0.0028)
0.7854
(0.0587)
-0.0077
(0.1288)
-0.0027
(0.0471)
0.0444
(0.0102)
0.4947
(0.0599)
-0.0069
(0.1261)
0.0034
(0.0268)
0.0162
(0.0055)
0.6016
(0.1015)
0.0174
(0.1797)
-0.0208
(0.0449)
0.0913
(0.0160)
0.2074
(0.1055)
-0.0450
(0.1457)
NONOIL
ln(h)
Implied X
Implied at
Implied up
0.1823
(0.0895)
0.0111
(0.0038)
0.6862
(0.0694)
0.2356
(0.1013)
INTER
ln(h)
Implied X
Implied a
Implied up
0.1101
(0.1305)
0.0118
(0.0045)
0.6906
(0.0799)
0.1335
(0.1428)
OECD
ln(h)
Implied X
Implied ax
Implied up
0.0864
(0.1551)
0.0187
(0.0077)
0.5416
(0.1426)
0.1062
(0.2027)
1152
QUARTERLYJOURNAL OF ECONOMICS
1153
1154
QUARTERLYJOURNAL OF ECONOMICS
Panel estimation permits us not only to allow for the individual country effects in the estimation of the other parameters of
the model, but also to get the estimates of these effects themselves.
In the case of LSDV estimation the recovery of the estimated
9. As they reported, ". . . the coefficient for human capital is insignificant and
enters with the wrong sign. . . . this result is independent of whether we use the
Kyriacou, Barro-Lee, or literacy data sets as proxies for the stock of human capital
in computing the growth rates of human capital" [Benhabib and Spiegel 1994,
p. 154].
1155
individual effects is direct because they are the estimated coefficients of the country dummies. If LSDV is implemented in the form
of "within-regression"-as
we do-the estimated country effects
are obtained as
(19)
pi =i
P3Xi-
9,
where
l
i=
TKiY
1 T-1
Yi,-1 =
X =
T
I
with t being the estimates of the time effects. In the case of the
correlated effects model, the MD estimation procedure yields
estimates of Kt. These can now be substituted back into equation
(14) to get the estimated country effects.
The estimated values of ,ui's are presented in column 3 of
Appendix 1.10The implied values of In A(O)i can be recovered from
the 'ii's using the formula, ki = (1 - e-XT) lnA(O)i. These are
presented in column 4. The dispersion and relative position of the
countries can be highlighted by computing the values of A(O)i and
expressing them relative to A(O)min(in the present case, the A(0) of
Somalia). These ratios may be called the A(0) index, and they are
presented in column 5. The rank of the countries in terms of this
index can be seen in column 6. The value of the index ranges from 1
to about 40, showing that the countries do vary enormously in
terms of their A(O) values. In general, however, we find a bottomheavy distribution. If we classify the countries according to whether
the estimated A(0) index is less than 5, between 5 and 10, 10 and
15, 15 and 20, 20 and 25, and greater than 25, and name the
corresponding groups for easy reference as I (Very Low), II (Low),
III (Medium), IV (High), V (Very High), and VI (Super High), then
we find that 63 (i.e., 66 percent) of the countries of the sample fall
into the lowest two groups. The histogram in Appendix 2 gives a
more visual display of the distribution along with Appendix other
information that makes it easy to read off from it both the rank and
the value of the A(0) index for particular countries.
The A(0) index is, obviously, a measure of efficiency with
which the countries are transforming their capital and labor
resources into output and hence is very close to the conventional
10. The estimates presented in Appendix 1 are based on NONOIL results. The
relative rankings of the countries do not change when these country effects are
estimated on the basis of the INTER sample. Also, the MD and LSDV yield similar
estimates when rounded up to two decimal points.
1156
QUARTERLYJOURNAL OF ECONOMICS
concept of total factor productivity (TFP). The important difference is that while the TFPs are computed for the individual
countries on the basis of their respective time series data, the
country effects are inherently based on cross-country comparison
and are not subsequent upon individualcountry-analysis.
Appendix 3 gives a two-way distribution of the countries in
terms of the A(O)index and continents. In general, it conforms to
what may be called the expected pattern. A large proportion (84
percent) of the Africancountries fall in the Very Low group. On the
other hand, most of the European countries are in the High and
Very High groups. Latin and CentralAmericancountries generally
tend to avoid the Very Low group, and instead belong mostly
(about half) to Group II. The Asian countries seem to be more
widely distributed across the groups. While a good number of them
(about one-third) fall into the Very Low group, many of the rest
make their way to the higher groups.
In recent years there has been renewed interest in in-depth
analysis of growth performance of individual countries (see, for
example, Young [1992, 1995]). In light of this interest, it may be
worthwhile to note certain "unexpected"or interesting aspects of
the results regardingthe A(O)index. First is the super high value of
this index for Hong Kong. Not only did it have the highest value,
but it surpassed by this measure the next ranking country
(Canada)by a wide margin. Hong Kong's super value of the A(O)
index, comparedwith that of Singapore, lends support to some of
Young's [1992] conclusions from his detailed analysis of the
comparativegrowth performanceof these two city states. Second,
the high values of the A(O)index for the United States and Canada
may to some extent dispel the thesis of productivity slowdown in
these countries, in particular, the United States. This is further
supported by the relatively low values of the A(O)index for both
Japan and Germany. Third, some of the countries did remarkably
well in terms of the A(O) index. Among these are, for example,
Israel, Trinidad and Tobago, and Venezuela and also to some
extent, Syria and South Africa.1"Fourth, equally noteworthy is the
poorA(O)index of some other countries. Among the latter are, for
example, Korea, Chile, Portugal, and India. This result, particularly regardingKoreaand Chile, may draw some attention.
Having seen the range, dispersion, and ranking, we may next
11. However, it may be necessary to check into the possible role of extraction of
oil and other mineral deposits in this regard, particularly with respect to countries
like Venezuela and South Africa.
1157
1y
- 7
ocb
0o
0 00
0
0
~~~~~~~~~
0000
000~~04J0
ln0
6"
%00
00~~~~~~~~~~~~~-1
OD0
I
I_
1ny8
ODOI&0
l~8
oec OP*
0 0
O~~~~~~~AOL
? ?
0c00
?
7IW94
FIGURE I
is
y(t)
((~g)1-k(~
where, as before, y(t) is per capita income at time t and k(t) is per
capita capital stock. The scatter plot matrix presented in Figure I
shows the relationship between the log of A(O)(denoted by ln aG)
and the log of y of 1960 and 1985 (denoted in the figure by lniy6
and lny85, respectively). It shows quite clearly that ln aO is very
1158
QUARTERLYJOURNAL OF ECONOMICS
closely associated with both in y6O and in y85. The simple correlation coefficients are 0.8656 and 0.9599 for they's of 1960 and 1985,
respectively. Note that the association between ln y60 and ln y85
themselves is very strong. This can be seen from the upper right or
lower left blocks of the scatter plot matrix. The simple correlation
coefficient between the two is 0.9202.12
Of course, one weakness of the above evidence is that we do not
have independent measures of A(0). The values of A(0) used to
determine the correlations are the outcome of an estimation
process in which values of y(t) themselves (along with other
variables) served as the data. Therefore, there may be an induced
element in the correlations cited above. However, even after
discounting for such a possibility, the strength of the correlations
probably remains very strong.
Next we turn to the issue of growth rates. Is the A(0) term
important in explaining growth? According to the Solow-CassKoopmans model, steady state growth is given by the exogenous
rate of technical progress. Hence, the focus here is on growth in
transition. Can A(0) affect transitional growth? Note that since
A(O) enters the production function in a multiplicative way, upon
log differencing of y(t) of any two time periods it would vanish. As
we can see from equation (20),
(21)
lny(t2) - lny(t2)
(1
0042
t1) +
1159
ln-aO
SGP
HKG
JPN
KOR
.05 -
BWA
OMRA
THA
PPT
ESP0
COG EGY
SYPINI
TP*AR
ChLPPYIRL
KCAN
J
SUP
MUSDOM LI
USA
MWI
.
XAUS
GSP
Z
E
LAMIO
L
YE
P
ZE
E
PS
G
NG N$p
SWODB
On
JW~~~~~~~~~~MPPK
Go
CD
CO0
CD
TZA
~~~~~MPT
CD
LSP
URY
FIGURL
mo~G
ZAP GHA
TCD
-.05
Plot
Scate
GflO
vesuEN68
FIGURE II
Scatter Plot of In aO versus GR6085
(22)
A()e
Yt~~
+ +
n +g + 6
OA/(-a)
Yo = A(0)aegt
(s/(n + g +))a/(
gk
a)
1160
line in the plot shows the fitted values from the bivariate regression
of GR6085 on In a0.13 This plot also helps to illustrate the relative
position of the particular countries in the In a0 scale and the
corresponding growth performance. (The codes to the abbreviated
country names used in the plot can be seen in Appendix 1.)
The above analysis therefore shows that higher values of A(0)
are associated not only with higher levels of per capita income, but
also with higher growth rates. The important question then is
what the determinants of A(0) are. In our discussion in Section II,
we have considered the qualitative list of the possible components
of A(0). The question, however, is whether we can determine (at
least approximately) their relative quantitative significance. This is
obviously a big question, and we cannot adequately address it
within the limits of the current paper. However, it indicates one of
the lines along which future research can proceed.14
Before concluding the discussion of this section, however, we
want to look into the relationship of A(0) with one other variable of
particular interest, namely, human capital. This has added relevance in view of our discussion in Section VII. We use two different
measures of human capital for this purpose. One is the SCHOOL
variable used by M-R-W and the other is Barro and Lee's HUMAN
variable. We have alluded to both these variables in our discussion
in the previous section. Recall that HUMAN in Barro and Lee is a
panel variable. For our current cross-sectional analysis, we compute averages from the data on HUMAN for different quinquennial
years of the respective countries. These can be seen in the last
column of Appendix 1. Figure III gives the scatter plot matrix of
ln a, SCHOOL, and HUMAN. The plot shows that both the
SCHOOL and HUMAN variables are indeed very closely related to
In aG. The positive association is closer to HUMAN than to
SCHOOL. Simple correlation coefficients of In a0 with SCHOOL
and HUMAN are 0.7120 and 0.7802, respectively. (The correlation
coefficient between SCHOOL and HUMAN is 0.7529.)
This correlation between In a0 with human capital variables
13. The regression has the following results:
GR6085
-7.5296
(1.1429)
+ 1.3166 lnaO;
(0.1594)
N = 96,
0.4145
F194 = 68.26.
The figures in parentheses are standard errors. This indicates that an increase in
In aO by 1, which in turn implies about a threefold increase in the value of A(0),
would be associated with a 1.32 percentage point increase in the growth rate.
14. The author is currently engaged in such research and intends to produce a
paper on this topic in future.
1161
9
15
school
0
scho0
0
a%
05
~~~~~~~0
0 0
~~~~~~
00
0 0
0510
0O~~~%
00
o~~~oR
o~~~~~~o~
~
8
0
9
~~
0~~~0
oo~~~,,o
10
-10
8 ,000070000
~~~ of SCOL
Oc~~~Q3~ Scatter~
0
0 Plo Mari
0
l0 a,
an5UA
0~~~~~~~~~~~~o
obtain
e
i
10
0
15
-0~0
ha
~~~~~~0
-5
51
1162
QUARTERLYJOURNAL OF ECONOMICS
this channel most likely runs through A. Of course, this does not
resolve the question quoted above, but it perhaps at least indicates
where to look for the answer.
IX. CONCLUSION
We argued for and implemented a panel data approach to the
study of cross-country growth, in particular, of the phenomenon of
convergence. This results in higher rates of conditional convergence and lower values of the elasticity of output with respect to
capital. These results are explainable in terms of correction for
omitted variable bias involved with the single cross-section regression. From growth theory's point of view, they highlight the
significance of the differences in the aggregate production functions.
However, the faster convergence that we observe is conditional. So to speak, this reveals the flip side of our finding. By being
more successful (through the panel framework) in controlling for
further sources of difference in the steady state levels of income, we
have, at the same time, made the obtained convergence hollower.
This is because convergence is more commonly understood as
different countries of the world approaching the same or similar
levels of income (i.e., in the "absolute" sense). There is probably
little solace to be derived from finding that countries in the world
are converging at a faster rate, when the points to which they are
converging remain very different.
This also sheds light on the issue of policy activism. The faster
rate of convergence may appear to reinforce the policy-irrelevance
ethos ascribed to the Solow model. In actuality, however, the
opposite is the case. Traditionally, only the saving and population
growth rates were thought to be the variables for policies to be
directed to. However, our study highlights the role of the A(O) term
as a determinant of the steady state level of income. It thus brings
to the fore the fact that, even with similar rates of saving and
population growth, a country can directly improve its long-run
economic position by bringing about improvements in the components of A(0). Also, improvements in A(O) can have salutary effects
on s and n leading to a further (indirect) increase in the steady state
level of income. Therefore, this study points to a richer scope for
policies in raising the long-run income levels of countries and in
quickening the pace of reaching them. One of the distinguishing
aspects of development economics as a branch of economics has
been its emphasis on and discussion of the elements that enter into
1163
A(0)j
Country
Code
Algeria
Angola
Benin
Botswana
Burundi
Cameroon
Ctrl.Afr.Rep.
Chad
Congo
Egypt
Ethiopia
Ghana
Ivory Coast
Kenya
Liberia
Madagascar
Malawi
Mali
Mauritania
Mauritius
Morocco
Mozambique
Niger
Nigeria
Rwanda
Senegal
Sierra Leone
DZA
AGO
BEN
BWA
BDI
CMR
CAR
TCD
COG
EGY
ETH
GHA
CIV
KEN
LBR
MDG
MWI
MLI
MRT
MUS
MAR
MOZ
NER
NGA
RWA
SEN
SLE
i
1.45
1.38
1.25
1.47
1.23
1.42
1.20
1.14
1.36
1.41
1.27
1.19
1.43
1.25
1.21
1.29
1.21
1.20
1.16
1.45
1.56
1.36
1.27
1.30
1.23
1.34
1.26
In A(0)i
A(0)min
Rank
SCHOOL
HUMAN
6.97
6.63
6.00
7.06
5.91
6.82
5.76
5.48
6.53
6.77
6.10
5.72
6.87
6.00
5.81
6.20
5.81
5.76
5.57
6.97
7.49
6.53
6.10
6.24
5.91
6.44
6.05
5.12
3.49
1.96
5.64
1.78
4.43
1.54
1.16
3.32
4.23
2.16
1.47
4.65
1.96
1.62
2.37
1.62
1.54
1.27
5.12
8.68
3.32
2.16
2.49
1.78
3.17
2.06
53
65
81
51
83
60
88
94
68
62
76
90
58
80
87
74
86
89
91
54
37
67
77
73
84
69
78
4.5
1.8
1.8
2.9
0.4
3.4
1.4
0.4
3.8
7.0
1.1
4.7
2.3
2.4
2.5
2.6
0.6
1.0
1.0
7.3
3.6
0.7
0.5
2.3
0.4
1.7
1.7
1.27
0.57*
1.96
1.75
2.71*
1.32*
1.93
1.76
1.04
2.04
0.47*
3.55
0.71
0.61
0.86*
1.81
1.17
1164
APPENDIX
(CONTINUED)
Country
Somalia
S. Africa
Sudan
Tanzania
Togo
Tunisia
Uganda
Zaire
Zambia
Zimbabwe
Bangladesh
Burma
Hong Kong
India
Israel
Japan
Jordan
S. Korea
Malaysia
Nepal
Pakistan
Philippines
Singapore
Sri Lanka
Syria
Thailand
Austria
Belgium
Denmark
Finland
France
Germany
Greece
Ireland
Italy
Holland
Norway
Portugal
Spain
Sweden
Switzerland
Turkey
U. Kingdom
Canada
Code
pi
In A(0)j
A(0)i
A(0)min
Rank
SCHOOL
HUMAN
SOM
ZAF
SDN
TZA
TGO
TUN
UGA
ZAR
ZMB
ZWE
BGD
BUR
HKG
IND
ISR
JPN
JOR
KOR
MYS
NPL
PAK
PHL
SGP
LKA
SYR
THA
AUT
BEL
DNK
FIN
FRA
DEU
GRC
IRL
ITA
NLD
NOR
PRT
ESP
SWE
CHE
TUR
GBR
CAN
1.11
1.60
1.22
1.15
1.25
1.53
1.33
1.15
1.14
1.33
1.38
1.29
1.89
1.25
1.70
1.75
1.52
1.60
1.60
1.36
1.46
1.45
1.77
1.41
1.64
1.51
1.72
1.75
1.74
1.66
1.75
1.72
1.60
1.60
1.69
1.73
1.77
1.58
1.75
1.73
1.70
1.53
1.73
1.81
5.33
7.69
5.86
5.52
6.00
7.35
6.39
5.52
5.48
6.39
6.63
6.20
9.08
6.00
8.17
8.41
7.30
7.69
7.69
6.53
7.01
6.97
8.50
6.77
7.88
7.25
8.26
8.41
8.36
7.97
8.41
8.26
7.69
7.69
8.12
8.31
8.50
7.59
8.41
8.31
8.17
7.35
8.31
8.69
1.00
10.52
1.70
1.21
1.96
7.89
2.88
1.21
1.16
2.88
3.66
2.37
38.49
1.96
17.01
21.63
7.17
10.03
10.52
3.32
5.37
5.12
23.81
4.23
12.75
6.83
18.73
21.63
20.61
14.04
21.63
18.73
10.52
10.52
16.21
19.65
23.81
9.56
21.63
19.65
17.09
7.52
23.82
28.85
96
28
85
93
82
41
72
92
95
71
63
75
1
74
17
10
44
33
31
66
52
55
6
61
25
47
15
8
11
23
9
14
30
21
19
12
5
34
7
13
18
43
4
2
1.1
3.0
2.0
0.5
2.9
4.3
1.1
3.6
2.4
4.4
3.2
3.5
7.2
5.1
9.5
10.9
10.8
10.2
7.3
2.3
3.0
10.6
9.0
8.3
8.8
4.4
8.0
9.3
10.7
11.5
8.9
8.4
7.9
11.4
7.1
10.7
10.0
5.8
8.0
7.9
4.8
5.5
8.9
10.6
4.46
0.50
2.02
0.96
1.28
1.36
1.32
2.70
2.08
1.21
1.27
5.85
2.19
8.04
7.39
2.46
5.65
3.74
0.33
1.46
5.12
3.72
4.68
2.24
3.81
5.42
8.16
9.82
8.58
5.21
8.15
5.63
6.96
5.34
7.23
8.19
2.34
4.53
8.25
7.49
2.35
7.92
9.16
1165
(CONTINUED)
Country
Code
pi
In A(0)i
A(0)i
A(0)min
Rank
SCHOOL
HUMAN
Costa Rica
Dom. Rep.
El Salvador
Guatemala
Haiti
Honduras
Jamaica
Mexico
Nicaragua
Panama
Trinidad
United States
Argentina
Bolivia
Brazil
Chile
Colombia
Ecuador
Paraguay
Peru
Uruguay
Venezuela
Australia
New Zealand
P. N. Guinea
CRI
DOM
SLV
GTM
HTL
HND
JAM
MEX
NIC
PAN
TTO
USA
ARG
BOL
BRA
CHL
COL
ECU
PRY
PER
URY
VEN
AUS
NZL
PNG
1.60
1.48
1.51
1.56
1.35
1.37
1.43
1.65
1.55
1.54
1.70
1.80
1.52
1.43
1.62
1.49
1.54
1.50
1.57
1.57
1.60
1.67
1.69
1.69
1.44
7.69
7.11
7.25
7.49
6.48
6.58
6.87
7.93
7.45
7.40
8.17
8.65
7.30
6.87
7.78
7.16
7.40
7.21
7.54
7.54
7.69
8.02
8.12
8.12
6.92
10.52
6.20
6.83
8.68
3.17
3.49
4.65
13.38
8.28
7.89
17.01
27.50
7.17
4.65
11.58
6.20
6.51
9.11
9.11
10.52
14.73
16.21
16.21
10.52
4.88
27
50
46
38
70
64
59
24
39
42
16
3
45
57
26
49
40
48
36
35
32
22
20
21
56
7.0
5.8
3.9
2.4
1.9
3.7
11.2
6.6
5.8
11.6
8.8
11.9
5.0
4.9
4.7
7.7
6.1
7.2
4.4
8.0
7.0
7.0
9.8
11.9
1.5
4.26
3.17
2.56
1.85
1.12
2.30
3.25
3.32
2.59
5.09
5.36
10.44
5.92
3.75
2.90
5.57
3.59
4.04
4.04
4.27
5.26
3.69
9.72
10.70
1.13
A period (.) indicates a missing value of the HUMAN variable, and an asterisk (*) indicates that the full set
of quinquennialvalues of the HUMANvariablewere not availablefor the countryand the averagehas been
computedon the basis of fewerdatapoints.
QUARTERLYJOURNAL OF ECONOMICS
1166
APPENDIX
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
PNG
BOL
CIV
JAM
CMR
LKA
EGY
BGD
HND
AGO
NPL
MOZ
COG
SEN
HTI
ZWE
UGA
NGA
MDG
BUR
ETH
NER
SLE
IND
KEN
BEN
TGO
BDI
RWA
SDN
MWI
LBR
CAF
MLI
GHA
MRT
ZAR
TZA
TCD
ZMB
SOM
(4.9)
(4.7)
(4.7)
(4.7)
(4.4)
(4.2)
(4.2)
(3.7)
(3.5)
(3.5)
(3.3)
(3.3)
(3.3)
(3.2)
(3.2)
(2.9)
(2.9)
(2.2)
(2.4)
(2.4)
(2.2)
(2.2)
(2.1)
(2.0)
(2.0)
(2.0)
(2.0)
(1.8)
(1.8)
(1.7)
(1.6)
(1.6)
(1.5)
(1.5)
(1.5)
(1.3)
(1.2)
(1.2)
(1.2)
(1.2)
(1.0)
1-5
PRT
PER
PRY
MAR
GTM
NIC
COL
TUN
PAN
TUR
JOR
ARG
SLV
THA
ECU
CHL
DOM
BWA
PAK
DZA
MUS
PHL
(9.6)
(9.1)
(9.1)
(8.9)
(8.7)
(8.3)
(7.9)
(7.9)
(7.9)
(7.5)
(7.2)
(7.2)
(6.8)
(6.8)
(6.5)
(6.2)
(6.2)
(5.6)
(5.4)
(5.1)
(5.1)
(5.1)
5-10
2:
HISTOGRAM OF THE
VEN
FIN
MEX
SYR
BRA
CRI
ZAF
IRL
GRC
MYS
URY
KOR
(14.7)
(14.0)
(13.4)
(12.8)
(11.6)
(10.5)
(10.5)
(10.5)
(10.5)
(10.5)
(10.5)
(10.0)
10-15
NLD
SWE
DEU
AUT
TTO
ISR
CHE
ITA
AUS
NZL
A(0) INDEX
(19.7)
(19.7)
(18.7)
(18.7)
(17.0)
(17.0)
(17.0)
(16.2)
(16.2)
(16.2)
15-20
GBR
NOR
SGP
ESP
BEL
FRA
JPN
DNK
(23.8)
(23.8)
(23.8)
(21.6)
(21.6)
(21.6) HKG (38.5)
(21.6) CAN (28.9)
(20.6) USA (27.5)
20-25
25+
The figures in parentheses are the corresponding values of the A() index.
The full names corresponding to the abbreviated names of the countries can be seen in Appendix 1.
1167
3:
DISTRIBUTION
A(0)
INDEX AND
CONTINENT
Continent
Group
I
Group
II
Group
III
Group
IV
Group
V
Group
VI
Total
Africa
Americaa
Asiab
Europec
31
4
6
0
5
11
4
2
1
5
3
3
0
1
3
6
0
0
2
6
0
2
1
0
37
23
19
17
Total
41
22
12
10
96
DISTANCE
ESTIMATION
APPENDIX4: NOTEON MINIMUM
The purpose of Appendix 4 is to give some details on the
Minimum Distance estimation method used in this paper. The
arguments for use of this method and its basic principles and
properties have been discussed in the text. Therefore, we concentrate here only on its operational aspects. Further details can be
seen in Chamberlain [1982, 1983].
The restricted model is
=it=
Yit-
+ Pxit + ki + Vito
where xi, is the notation for [in (s) - in (n + g + r)]. We ignore the
time effect term 9t which is taken care of by appropriate time
dummies.
This model is augmented by the following specifications for pi
andyio:
i=
+
K0Xij
Ko
YiO = ko +
+ K2Xi2 +
kXil + 4)2Xi2 +
+ KTXiT+ PIi
+ 4)TXiT + hi,
0.
Y1=
QUARTERLYJOURNAL OF ECONOMICS
1168
y3 =2pX
Y4
Y5
Syj3
B
Y213
ly3
y41
0
3
0
0
?0
0
_y3
y2
13
_y
y3r
_y2r
x1
x2
x3
x4
y4
_y1
p
X5
y5
Y3 Y0
1
1+ y
U1
U2
+ Ut3
u4
U5
l+ y+ y2
1+y+y2+y3
1+y+y2+y3+y4
IP
(3 0
-y3
~2p _ p
'y$3 _y3
p~2 _p
0 0
0 0
~2
0 +
03
y
y
1+y
1 +l+ y++ y2 + y3
y y2 )K.y
+ y + y2 + y3 + y4
If xe's are strictly exogenous, then they are uncorrelated with the
1169
-l(x'x/
l'],
where HOis the matrix of true coefficients and Fx = E(x ixj). It may
be noted that ft is a heteroskedasticity-consistent weighting matrix. In actual implementation we need to use its consistent sample
analog,
Ni=
z[(Yi
- Ji'xj)(Yi-
ix)' 0 S;-(xixltS;1],
REFERENCES
Amemiya, Takeshi, "A Note on the Estimation of Balestra-Nerlove Models,"
Technical Report No. 4, Institute for Mathematical Studies in Social Sciences,
Stanford University, 1967.
Barro, Robert J., "Economic Growth in a Cross Section of Countries," NBER
Working Paper No. 3120, Cambridge, September, 1989.
Barro, Robert J., and Jong-Wha Lee, "International Comparisons of Educational
Attainment," Journal of Monetary Economics, XXXII (1993), 363-94.
Barro, Robert J., and Xavier Sala-i-Martin, "Convergence," Journal of Political
Economy, C (1992),223-51.
Baumol, William J., "Productivity Growth, Convergence and Welfare: What the
Long-Run Data Show?" American Economic Review, LXXVI (1986), 1072-85.
Benhabib Jess, and Mark M. Spiegel, "The Role of Human Capital in Economic
Development: Evidence from Aggregate Cross-Country Data," Journal of
Monetary Economics, XXXIV (1994), 143-73.
1170
QUARTERLYJOURNAL OF ECONOMICS